Book a 20-min call
AI integration · fixed price · 5 working days

Your AI assistant can't see your business data. I fix that.

Every AI pilot demos beautifully and then dies on the same wall: the model has no access to the ERP, the CRM, or the database where the actual answers live. I build the connector that closes that gap — the same way I do it in production, for companies running Sage 100, Salesforce and Microsoft 365 every day.

Book a free 20-min call See exactly what you get

Start with a €290 proof of concept on your own data — 48 hours. Credited against the full build if you go ahead.

The problem isn't the model. It's the plumbing.

You don't need a better LLM. You need the one you already have to read live stock levels, open invoices, order history and pipeline — from the systems that actually hold them. That's an integration problem, and it's where most AI projects quietly stall:

01

The data is in a system built in 1998

Sage 100, Macola, an on-prem SQL Server, a Windows box under someone's desk. No clean REST API, no documentation, and a vendor who charges five figures to look at it.

02

It works in the demo, not on Tuesday

A prototype that answers three questions correctly is not a system. Production means auth, rate limits, permissions, error states, and being right when the data is messy.

03

Nobody wants to own it

Your IT team is at capacity, the AI consultancy wants a six-figure discovery phase, and the freelancer who built the prototype has moved on.

04

Security says no

And they're right to — until someone scopes read-only access properly, keeps credentials out of the model's context, and can show exactly what the assistant can and cannot touch.

What you get, in five working days

A working connector between your AI assistant and your real systems — delivered, deployed, documented, and demonstrated on your own data. Not a slide deck, not a proof of concept that needs another phase.

Included

  • An MCP server exposing your data to Claude, ChatGPT or your own app
  • Read-only access, scoped to exactly the tables and fields you approve
  • Authentication handled properly — credentials never reach the model
  • A working assistant answering real questions on your live data
  • Deployment to your infrastructure or a hosted endpoint you own
  • A short handover doc so your team can extend it
  • 30 days of bug fixes after delivery

Not included

  • Write access on day one — reads first, writes once you trust it
  • Model fine-tuning (you almost certainly don't need it)
  • Replacing your ERP, CRM or reporting stack
  • A twelve-week discovery phase
  • Anything I can't demo working by Friday
Sage 100SalesforceMicrosoft 365 / Graph TableauSQL ServerPostgres / MySQL REST / SOAPMCPPythonTypeScript

Why me

I build and run AI systems that are wired into live business data — not demos. In production right now: assistants integrated with Sage 100, Salesforce, Tableau and Microsoft 365 across a multi-company portfolio, exposing 75+ tools through a single hosted connector, used daily by supply chain, finance, sales and customer service teams across dozens of staff.

That means I've already hit the walls you're about to hit: the ERP with no real API, the read-only credential that isn't actually read-only, the connector that works on your laptop and dies behind the corporate proxy, the assistant that confidently invents a stock number. Solving those is the job. The AI part is the easy part.

Two ways to start

You don't know me yet, so don't take a big swing first. Start with the small one — it's designed to prove I can actually do this on your system, in two days, for less than the cost of a day of most consultants' time.

Start here
€290proof of concept — 48 hours
  • I connect to your system read-only and get one real question answered correctly on your live data
  • You get a short written scope: what a full integration takes, what it costs, what could go wrong
  • Delivered in 48 hours, on a call, on your data — not a slide deck
  • If you go ahead with the full build, the €290 comes off the price
  • If I find it can't be done sensibly, I tell you and you've spent €290 instead of €15,000
€1,490full connector — five working days
  • Everything in the scope above, built, deployed, documented and handed over
  • 50% to start, 50% on delivery — you only pay the balance if it works
  • Acceptance is three questions you choose, answered correctly on your data
  • 30 days of fixes after delivery
  • Optional: €390/month to keep it running, monitored and extended
€3,900per month — integration partner
  • For companies rolling AI out across several systems rather than one
  • Continuous build: new connectors, new tools, new data sources each month
  • You own everything produced; no lock-in, cancel with 30 days' notice
  • Priority response, direct line, no ticket queue

For comparison: a consultancy will quote €15,000–50,000 and a three-month discovery phase for the first of these. The difference isn't quality — it's that they have a sales team, three layers of management, and an incentive to make the project bigger. Invoiced through a registered company; EU VAT handled.

How it runs

DAY 0

20-minute call

You tell me which system holds the data and what questions you want answered. I tell you honestly whether five days is realistic. Free.

DAY 1

Access & scope

Read-only credentials, a written scope, and the three questions your assistant must answer correctly to count as done.

DAY 2–4

Build

Connector, auth, tool definitions, error handling. You get a working endpoint to poke at before it's finished.

DAY 5

Demo & handover

Live demo on your data against the agreed questions, deployed, documented, handed over. Then 30 days of fixes.

Questions

Our data is in an old system with no API. Can you still do it?

Usually yes — that's the normal case, not the exception. Direct database reads, vendor connectors, ODBC, file exports on a schedule. I'll tell you on the call if it's genuinely not possible, before you pay anything.

Is our data safe?

Read-only by default, scoped to fields you approve. Credentials live in your environment, never in the model's context. You get a written list of exactly what the assistant can reach. Happy to work inside your VPN or on your own cloud account.

What if it isn't done in five days?

Then you don't pay the balance. The scope is deliberately small enough that this doesn't happen — one system, agreed questions. If you want three systems, that's three engagements, priced up front.

Do we need a specific AI vendor?

No. The connector speaks MCP, so it works with Claude, ChatGPT and most agent frameworks, and you can switch models later without rebuilding the integration.

Can you work with our existing developers?

Yes, and it's usually faster. I hand over readable code and a doc written for the person who inherits it.

Start with a 20-minute call

Tell me the system, the questions you want answered, and your timeline. If I can't do it in five days I'll say so on the call — and if I know someone better suited, I'll point you there.

Or message me on LinkedIn. Based in Sarajevo, working across EU and US time zones. Invoicing through a registered company.